Automatic merge of rsync://rsync.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6.git

20 files changed, 1314 insertions, 122 deletions

diff --git a/include/asm-ia64/hw_irq.h b/include/asm-ia64/hw_irq.h
index 041ab8c51a6..cd4e06b74ab 100644
--- a/include/asm-ia64/hw_irq.h
+++ b/include/asm-ia64/hw_irq.h
@@ -81,6 +81,7 @@ extern __u8 isa_irq_to_vector_map[16];
 
 extern struct hw_interrupt_type irq_type_ia64_lsapic;	/* CPU-internal interrupt controller */
 
+extern int assign_irq_vector_nopanic (int irq); /* allocate a free vector without panic */
 extern int assign_irq_vector (int irq);	/* allocate a free vector */
 extern void free_irq_vector (int vector);
 extern void ia64_send_ipi (int cpu, int vector, int delivery_mode, int redirect);
diff --git a/include/asm-ia64/pal.h b/include/asm-ia64/pal.h
index 5dd477ffb88..2303a10ee59 100644
--- a/include/asm-ia64/pal.h
+++ b/include/asm-ia64/pal.h
@@ -67,6 +67,7 @@
 #define PAL_REGISTER_INFO	39	/* return AR and CR register information*/
 #define PAL_SHUTDOWN		40	/* enter processor shutdown state */
 #define PAL_PREFETCH_VISIBILITY	41	/* Make Processor Prefetches Visible */
+#define PAL_LOGICAL_TO_PHYSICAL 42	/* returns information on logical to physical processor mapping */
 
 #define PAL_COPY_PAL		256	/* relocate PAL procedures and PAL PMI */
 #define PAL_HALT_INFO		257	/* return the low power capabilities of processor */
@@ -1559,6 +1560,73 @@ ia64_pal_prefetch_visibility (s64 trans_type)
 	return iprv.status;
 }
 
+/* data structure for getting information on logical to physical mappings */
+typedef union pal_log_overview_u {
+	struct {
+		u64	num_log		:16,	/* Total number of logical
+						 * processors on this die
+						 */
+			tpc		:8,	/* Threads per core */
+			reserved3	:8,	/* Reserved */
+			cpp		:8,	/* Cores per processor */
+			reserved2	:8,	/* Reserved */
+			ppid		:8,	/* Physical processor ID */
+			reserved1	:8;	/* Reserved */
+	} overview_bits;
+	u64 overview_data;
+} pal_log_overview_t;
+
+typedef union pal_proc_n_log_info1_u{
+	struct {
+		u64	tid		:16,	/* Thread id */
+			reserved2	:16,	/* Reserved */
+			cid		:16,	/* Core id */
+			reserved1	:16;	/* Reserved */
+	} ppli1_bits;
+	u64	ppli1_data;
+} pal_proc_n_log_info1_t;
+
+typedef union pal_proc_n_log_info2_u {
+	struct {
+		u64	la		:16,	/* Logical address */
+			reserved	:48;	/* Reserved */
+	} ppli2_bits;
+	u64	ppli2_data;
+} pal_proc_n_log_info2_t;
+
+typedef struct pal_logical_to_physical_s
+{
+	pal_log_overview_t overview;
+	pal_proc_n_log_info1_t ppli1;
+	pal_proc_n_log_info2_t ppli2;
+} pal_logical_to_physical_t;
+
+#define overview_num_log	overview.overview_bits.num_log
+#define overview_tpc		overview.overview_bits.tpc
+#define overview_cpp		overview.overview_bits.cpp
+#define overview_ppid		overview.overview_bits.ppid
+#define log1_tid		ppli1.ppli1_bits.tid
+#define log1_cid		ppli1.ppli1_bits.cid
+#define log2_la			ppli2.ppli2_bits.la
+
+/* Get information on logical to physical processor mappings. */
+static inline s64
+ia64_pal_logical_to_phys(u64 proc_number, pal_logical_to_physical_t *mapping)
+{
+	struct ia64_pal_retval iprv;
+
+	PAL_CALL(iprv, PAL_LOGICAL_TO_PHYSICAL, proc_number, 0, 0);
+
+	if (iprv.status == PAL_STATUS_SUCCESS)
+	{
+		if (proc_number == 0)
+			mapping->overview.overview_data = iprv.v0;
+		mapping->ppli1.ppli1_data = iprv.v1;
+		mapping->ppli2.ppli2_data = iprv.v2;
+	}
+
+	return iprv.status;
+}
 #endif /* __ASSEMBLY__ */
 
 #endif /* _ASM_IA64_PAL_H */
diff --git a/include/asm-ia64/perfmon.h b/include/asm-ia64/perfmon.h
index 136c60e6bfc..ed5416c5b1a 100644
--- a/include/asm-ia64/perfmon.h
+++ b/include/asm-ia64/perfmon.h
@@ -254,6 +254,18 @@ extern int pfm_mod_write_dbrs(struct task_struct *task, void *req, unsigned int
 #define PFM_CPUINFO_DCR_PP	0x2	/* if set the system wide session has started */
 #define PFM_CPUINFO_EXCL_IDLE	0x4	/* the system wide session excludes the idle task */
 
+/*
+ * sysctl control structure. visible to sampling formats
+ */
+typedef struct {
+	int	debug;		/* turn on/off debugging via syslog */
+	int	debug_ovfl;	/* turn on/off debug printk in overflow handler */
+	int	fastctxsw;	/* turn on/off fast (unsecure) ctxsw */
+	int	expert_mode;	/* turn on/off value checking */
+} pfm_sysctl_t;
+extern pfm_sysctl_t pfm_sysctl;
+
+
 #endif /* __KERNEL__ */
 
 #endif /* _ASM_IA64_PERFMON_H */
diff --git a/include/asm-ia64/pgalloc.h b/include/asm-ia64/pgalloc.h
index 0f05dc8bd46..a5f214554af 100644
--- a/include/asm-ia64/pgalloc.h
+++ b/include/asm-ia64/pgalloc.h
@@ -22,146 +22,124 @@
 
 #include <asm/mmu_context.h>
 
-/*
- * Very stupidly, we used to get new pgd's and pmd's, init their contents
- * to point to the NULL versions of the next level page table, later on
- * completely re-init them the same way, then free them up.  This wasted
- * a lot of work and caused unnecessary memory traffic.  How broken...
- * We fix this by caching them.
- */
-#define pgd_quicklist		(local_cpu_data->pgd_quick)
-#define pmd_quicklist		(local_cpu_data->pmd_quick)
-#define pgtable_cache_size	(local_cpu_data->pgtable_cache_sz)
+DECLARE_PER_CPU(unsigned long *, __pgtable_quicklist);
+#define pgtable_quicklist __ia64_per_cpu_var(__pgtable_quicklist)
+DECLARE_PER_CPU(long, __pgtable_quicklist_size);
+#define pgtable_quicklist_size __ia64_per_cpu_var(__pgtable_quicklist_size)
 
-static inline pgd_t*
-pgd_alloc_one_fast (struct mm_struct *mm)
+static inline long pgtable_quicklist_total_size(void)
+{
+	long ql_size = 0;
+	int cpuid;
+
+	for_each_online_cpu(cpuid) {
+		ql_size += per_cpu(__pgtable_quicklist_size, cpuid);
+	}
+	return ql_size;
+}
+
+static inline void *pgtable_quicklist_alloc(void)
 {
 	unsigned long *ret = NULL;
 
 	preempt_disable();
 
-	ret = pgd_quicklist;
+	ret = pgtable_quicklist;
 	if (likely(ret != NULL)) {
-		pgd_quicklist = (unsigned long *)(*ret);
+		pgtable_quicklist = (unsigned long *)(*ret);
 		ret[0] = 0;
-		--pgtable_cache_size;
-	} else
-		ret = NULL;
-
-	preempt_enable();
+		--pgtable_quicklist_size;
+		preempt_enable();
+	} else {
+		preempt_enable();
+		ret = (unsigned long *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
+	}
 
-	return (pgd_t *) ret;
+	return ret;
 }
 
-static inline pgd_t*
-pgd_alloc (struct mm_struct *mm)
+static inline void pgtable_quicklist_free(void *pgtable_entry)
 {
-	/* the VM system never calls pgd_alloc_one_fast(), so we do it here. */
-	pgd_t *pgd = pgd_alloc_one_fast(mm);
+#ifdef CONFIG_NUMA
+	unsigned long nid = page_to_nid(virt_to_page(pgtable_entry));
 
-	if (unlikely(pgd == NULL)) {
-		pgd = (pgd_t *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
+	if (unlikely(nid != numa_node_id())) {
+		free_page((unsigned long)pgtable_entry);
+		return;
 	}
-	return pgd;
-}
+#endif
 
-static inline void
-pgd_free (pgd_t *pgd)
-{
 	preempt_disable();
-	*(unsigned long *)pgd = (unsigned long) pgd_quicklist;
-	pgd_quicklist = (unsigned long *) pgd;
-	++pgtable_cache_size;
+	*(unsigned long *)pgtable_entry = (unsigned long)pgtable_quicklist;
+	pgtable_quicklist = (unsigned long *)pgtable_entry;
+	++pgtable_quicklist_size;
 	preempt_enable();
 }
 
-static inline void
-pud_populate (struct mm_struct *mm, pud_t *pud_entry, pmd_t *pmd)
+static inline pgd_t *pgd_alloc(struct mm_struct *mm)
 {
-	pud_val(*pud_entry) = __pa(pmd);
+	return pgtable_quicklist_alloc();
 }
 
-static inline pmd_t*
-pmd_alloc_one_fast (struct mm_struct *mm, unsigned long addr)
+static inline void pgd_free(pgd_t * pgd)
 {
-	unsigned long *ret = NULL;
-
-	preempt_disable();
-
-	ret = (unsigned long *)pmd_quicklist;
-	if (likely(ret != NULL)) {
-		pmd_quicklist = (unsigned long *)(*ret);
-		ret[0] = 0;
-		--pgtable_cache_size;
-	}
-
-	preempt_enable();
-
-	return (pmd_t *)ret;
+	pgtable_quicklist_free(pgd);
 }
 
-static inline pmd_t*
-pmd_alloc_one (struct mm_struct *mm, unsigned long addr)
+static inline void
+pud_populate(struct mm_struct *mm, pud_t * pud_entry, pmd_t * pmd)
 {
-	pmd_t *pmd = (pmd_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
+	pud_val(*pud_entry) = __pa(pmd);
+}
 
-	return pmd;
+static inline pmd_t *pmd_alloc_one(struct mm_struct *mm, unsigned long addr)
+{
+	return pgtable_quicklist_alloc();
 }
 
-static inline void
-pmd_free (pmd_t *pmd)
+static inline void pmd_free(pmd_t * pmd)
 {
-	preempt_disable();
-	*(unsigned long *)pmd = (unsigned long) pmd_quicklist;
-	pmd_quicklist = (unsigned long *) pmd;
-	++pgtable_cache_size;
-	preempt_enable();
+	pgtable_quicklist_free(pmd);
 }
 
 #define __pmd_free_tlb(tlb, pmd)	pmd_free(pmd)
 
 static inline void
-pmd_populate (struct mm_struct *mm, pmd_t *pmd_entry, struct page *pte)
+pmd_populate(struct mm_struct *mm, pmd_t * pmd_entry, struct page *pte)
 {
 	pmd_val(*pmd_entry) = page_to_phys(pte);
 }
 
 static inline void
-pmd_populate_kernel (struct mm_struct *mm, pmd_t *pmd_entry, pte_t *pte)
+pmd_populate_kernel(struct mm_struct *mm, pmd_t * pmd_entry, pte_t * pte)
 {
 	pmd_val(*pmd_entry) = __pa(pte);
 }
 
-static inline struct page *
-pte_alloc_one (struct mm_struct *mm, unsigned long addr)
+static inline struct page *pte_alloc_one(struct mm_struct *mm,
+					 unsigned long addr)
 {
-	struct page *pte = alloc_pages(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO, 0);
-
-	return pte;
+	return virt_to_page(pgtable_quicklist_alloc());
 }
 
-static inline pte_t *
-pte_alloc_one_kernel (struct mm_struct *mm, unsigned long addr)
+static inline pte_t *pte_alloc_one_kernel(struct mm_struct *mm,
+					  unsigned long addr)
 {
-	pte_t *pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
-
-	return pte;
+	return pgtable_quicklist_alloc();
 }
 
-static inline void
-pte_free (struct page *pte)
+static inline void pte_free(struct page *pte)
 {
-	__free_page(pte);
+	pgtable_quicklist_free(page_address(pte));
 }
 
-static inline void
-pte_free_kernel (pte_t *pte)
+static inline void pte_free_kernel(pte_t * pte)
 {
-	free_page((unsigned long) pte);
+	pgtable_quicklist_free(pte);
 }
 
-#define __pte_free_tlb(tlb, pte)	tlb_remove_page((tlb), (pte))
+#define __pte_free_tlb(tlb, pte)	pte_free(pte)
 
-extern void check_pgt_cache (void);
+extern void check_pgt_cache(void);
 
-#endif /* _ASM_IA64_PGALLOC_H */
+#endif				/* _ASM_IA64_PGALLOC_H */
diff --git a/include/asm-ia64/processor.h b/include/asm-ia64/processor.h
index 2807f8d766d..9e1ba8b7fb6 100644
--- a/include/asm-ia64/processor.h
+++ b/include/asm-ia64/processor.h
@@ -137,9 +137,6 @@ struct cpuinfo_ia64 {
 	__u64 nsec_per_cyc;	/* (1000000000<<IA64_NSEC_PER_CYC_SHIFT)/itc_freq */
 	__u64 unimpl_va_mask;	/* mask of unimplemented virtual address bits (from PAL) */
 	__u64 unimpl_pa_mask;	/* mask of unimplemented physical address bits (from PAL) */
-	__u64 *pgd_quick;
-	__u64 *pmd_quick;
-	__u64 pgtable_cache_sz;
 	__u64 itc_freq;		/* frequency of ITC counter */
 	__u64 proc_freq;	/* frequency of processor */
 	__u64 cyc_per_usec;	/* itc_freq/1000000 */
@@ -151,6 +148,13 @@ struct cpuinfo_ia64 {
 #ifdef CONFIG_SMP
 	__u64 loops_per_jiffy;
 	int cpu;
+	__u32 socket_id;	/* physical processor socket id */
+	__u16 core_id;		/* core id */
+	__u16 thread_id;	/* thread id */
+	__u16 num_log;		/* Total number of logical processors on
+				 * this socket that were successfully booted */
+	__u8  cores_per_socket;	/* Cores per processor socket */
+	__u8  threads_per_core;	/* Threads per core */
 #endif
 
 	/* CPUID-derived information: */
diff --git a/include/asm-ia64/sal.h b/include/asm-ia64/sal.h
index 240676f7539..29df88bdd2b 100644
--- a/include/asm-ia64/sal.h
+++ b/include/asm-ia64/sal.h
@@ -91,6 +91,7 @@ extern spinlock_t sal_lock;
 #define SAL_PCI_CONFIG_READ		0x01000010
 #define SAL_PCI_CONFIG_WRITE		0x01000011
 #define SAL_FREQ_BASE			0x01000012
+#define SAL_PHYSICAL_ID_INFO		0x01000013
 
 #define SAL_UPDATE_PAL			0x01000020
 
@@ -815,6 +816,17 @@ ia64_sal_update_pal (u64 param_buf, u64 scratch_buf, u64 scratch_buf_size,
 	return isrv.status;
 }
 
+/* Get physical processor die mapping in the platform. */
+static inline s64
+ia64_sal_physical_id_info(u16 *splid)
+{
+	struct ia64_sal_retval isrv;
+	SAL_CALL(isrv, SAL_PHYSICAL_ID_INFO, 0, 0, 0, 0, 0, 0, 0);
+	if (splid)
+		*splid = isrv.v0;
+	return isrv.status;
+}
+
 extern unsigned long sal_platform_features;
 
 extern int (*salinfo_platform_oemdata)(const u8 *, u8 **, u64 *);
diff --git a/include/asm-ia64/smp.h b/include/asm-ia64/smp.h
index c4a227acfeb..3ba1a061e4a 100644
--- a/include/asm-ia64/smp.h
+++ b/include/asm-ia64/smp.h
@@ -56,6 +56,10 @@ extern struct smp_boot_data {
 extern char no_int_routing __devinitdata;
 
 extern cpumask_t cpu_online_map;
+extern cpumask_t cpu_core_map[NR_CPUS];
+extern cpumask_t cpu_sibling_map[NR_CPUS];
+extern int smp_num_siblings;
+extern int smp_num_cpucores;
 extern void __iomem *ipi_base_addr;
 extern unsigned char smp_int_redirect;
 
@@ -124,6 +128,7 @@ extern int smp_call_function_single (int cpuid, void (*func) (void *info), void
 extern void smp_send_reschedule (int cpu);
 extern void lock_ipi_calllock(void);
 extern void unlock_ipi_calllock(void);
+extern void identify_siblings (struct cpuinfo_ia64 *);
 
 #else
 
diff --git a/include/asm-ia64/sn/addrs.h b/include/asm-ia64/sn/addrs.h
index c916bd22767..960d626ee58 100644
--- a/include/asm-ia64/sn/addrs.h
+++ b/include/asm-ia64/sn/addrs.h
@@ -154,8 +154,9 @@
  *           the chiplet id is zero.  If we implement TIO-TIO dma, we might need
  *           to insert a chiplet id into this macro.  However, it is our belief
  *           right now that this chiplet id will be ICE, which is also zero.
+ *           Nasid starts on bit 40.
  */
-#define PHYS_TO_TIODMA(x)	( (((u64)(x) & NASID_MASK) << 2) | NODE_OFFSET(x))
+#define PHYS_TO_TIODMA(x)	( (((u64)(NASID_GET(x))) << 40) | NODE_OFFSET(x))
 #define PHYS_TO_DMA(x)          ( (((u64)(x) & NASID_MASK) >> 2) | NODE_OFFSET(x))
 
 
@@ -168,7 +169,10 @@
 #define TIO_BWIN_SIZE_BITS		30	/* big window size: 1G */
 #define NODE_SWIN_BASE(n, w)		((w == 0) ? NODE_BWIN_BASE((n), SWIN0_BIGWIN) \
 		: RAW_NODE_SWIN_BASE(n, w))
+#define TIO_SWIN_BASE(n, w) 		(TIO_IO_BASE(n) + \
+					    ((u64) (w) << TIO_SWIN_SIZE_BITS))
 #define NODE_IO_BASE(n)			(GLOBAL_MMR_SPACE | NASID_SPACE(n))
+#define TIO_IO_BASE(n)                  (UNCACHED | NASID_SPACE(n))
 #define BWIN_SIZE			(1UL << BWIN_SIZE_BITS)
 #define NODE_BWIN_BASE0(n)		(NODE_IO_BASE(n) + BWIN_SIZE)
 #define NODE_BWIN_BASE(n, w)		(NODE_BWIN_BASE0(n) + ((u64) (w) << BWIN_SIZE_BITS))
diff --git a/include/asm-ia64/sn/bte.h b/include/asm-ia64/sn/bte.h
index 0ec27f99c18..f50da3d91d0 100644
--- a/include/asm-ia64/sn/bte.h
+++ b/include/asm-ia64/sn/bte.h
@@ -3,7 +3,7 @@
  * License.  See the file "COPYING" in the main directory of this archive
  * for more details.
  *
- * Copyright (c) 2000-2004 Silicon Graphics, Inc.  All Rights Reserved.
+ * Copyright (c) 2000-2005 Silicon Graphics, Inc.  All Rights Reserved.
  */
 
 
@@ -13,8 +13,12 @@
 #include <linux/timer.h>
 #include <linux/spinlock.h>
 #include <linux/cache.h>
+#include <asm/sn/pda.h>
 #include <asm/sn/types.h>
+#include <asm/sn/shub_mmr.h>
 
+#define IBCT_NOTIFY             (0x1UL << 4)
+#define IBCT_ZFIL_MODE          (0x1UL << 0)
 
 /* #define BTE_DEBUG */
 /* #define BTE_DEBUG_VERBOSE */
@@ -39,8 +43,36 @@
 
 
 /* Define hardware */
-#define BTES_PER_NODE 2
+#define BTES_PER_NODE (is_shub2() ? 4 : 2)
+#define MAX_BTES_PER_NODE 4
 
+#define BTE2OFF_CTRL	(0)
+#define BTE2OFF_SRC	(SH2_BT_ENG_SRC_ADDR_0 - SH2_BT_ENG_CSR_0)
+#define BTE2OFF_DEST	(SH2_BT_ENG_DEST_ADDR_0 - SH2_BT_ENG_CSR_0)
+#define BTE2OFF_NOTIFY	(SH2_BT_ENG_NOTIF_ADDR_0 - SH2_BT_ENG_CSR_0)
+
+#define BTE_BASE_ADDR(interface) 				\
+    (is_shub2() ? (interface == 0) ? SH2_BT_ENG_CSR_0 :		\
+		  (interface == 1) ? SH2_BT_ENG_CSR_1 :		\
+		  (interface == 2) ? SH2_BT_ENG_CSR_2 :		\
+		  		     SH2_BT_ENG_CSR_3 		\
+		: (interface == 0) ? IIO_IBLS0 : IIO_IBLS1)
+
+#define BTE_SOURCE_ADDR(base)					\
+    (is_shub2() ? base + (BTE2OFF_SRC/8) 			\
+		: base + (BTEOFF_SRC/8))
+
+#define BTE_DEST_ADDR(base)					\
+    (is_shub2() ? base + (BTE2OFF_DEST/8) 			\
+		: base + (BTEOFF_DEST/8))
+
+#define BTE_CTRL_ADDR(base)					\
+    (is_shub2() ? base + (BTE2OFF_CTRL/8) 			\
+		: base + (BTEOFF_CTRL/8))
+
+#define BTE_NOTIF_ADDR(base)					\
+    (is_shub2() ? base + (BTE2OFF_NOTIFY/8) 			\
+		: base + (BTEOFF_NOTIFY/8))
 
 /* Define hardware modes */
 #define BTE_NOTIFY (IBCT_NOTIFY)
@@ -68,14 +100,18 @@
 #define BTE_LNSTAT_STORE(_bte, _x)					\
 			HUB_S(_bte->bte_base_addr, (_x))
 #define BTE_SRC_STORE(_bte, _x)						\
-			HUB_S(_bte->bte_base_addr + (BTEOFF_SRC/8), (_x))
+			HUB_S(_bte->bte_source_addr, (_x))
 #define BTE_DEST_STORE(_bte, _x)					\
-			HUB_S(_bte->bte_base_addr + (BTEOFF_DEST/8), (_x))
+			HUB_S(_bte->bte_destination_addr, (_x))
 #define BTE_CTRL_STORE(_bte, _x)					\
-			HUB_S(_bte->bte_base_addr + (BTEOFF_CTRL/8), (_x))
+			HUB_S(_bte->bte_control_addr, (_x))
 #define BTE_NOTIF_STORE(_bte, _x)					\
-			HUB_S(_bte->bte_base_addr + (BTEOFF_NOTIFY/8), (_x))
+			HUB_S(_bte->bte_notify_addr, (_x))
 
+#define BTE_START_TRANSFER(_bte, _len, _mode)				\
+	is_shub2() ? BTE_CTRL_STORE(_bte, IBLS_BUSY | (_mode << 24) | _len) \
+		: BTE_LNSTAT_STORE(_bte, _len);				\
+		  BTE_CTRL_STORE(_bte, _mode)
 
 /* Possible results from bte_copy and bte_unaligned_copy */
 /* The following error codes map into the BTE hardware codes
@@ -110,6 +146,10 @@ typedef enum {
 struct bteinfo_s {
 	volatile u64 notify ____cacheline_aligned;
 	u64 *bte_base_addr ____cacheline_aligned;
+	u64 *bte_source_addr;
+	u64 *bte_destination_addr;
+	u64 *bte_control_addr;
+	u64 *bte_notify_addr;
 	spinlock_t spinlock;
 	cnodeid_t bte_cnode;	/* cnode                            */
 	int bte_error_count;	/* Number of errors encountered     */
@@ -117,6 +157,7 @@ struct bteinfo_s {
 	int cleanup_active;	/* Interface is locked for cleanup  */
 	volatile bte_result_t bh_error;	/* error while processing   */
 	volatile u64 *most_rcnt_na;
+	struct bteinfo_s *btes_to_try[MAX_BTES_PER_NODE];
 };
 
 
diff --git a/include/asm-ia64/sn/geo.h b/include/asm-ia64/sn/geo.h
index f566343d25f..84b254603b8 100644
--- a/include/asm-ia64/sn/geo.h
+++ b/include/asm-ia64/sn/geo.h
@@ -18,32 +18,34 @@
 #define GEOID_SIZE	8	/* Would 16 be better?  The size can
 				   be different on different platforms. */
 
-#define MAX_SLABS	0xe	/* slabs per module */
+#define MAX_SLOTS	0xf	/* slots per module */
+#define MAX_SLABS	0xf	/* slabs per slot */
 
 typedef unsigned char	geo_type_t;
 
 /* Fields common to all substructures */
-typedef struct geo_any_s {
+typedef struct geo_common_s {
     moduleid_t	module;		/* The module (box) this h/w lives in */
     geo_type_t	type;		/* What type of h/w is named by this geoid_t */
-    slabid_t	slab;		/* The logical assembly within the module */
-} geo_any_t;
+    slabid_t	slab:4;		/* slab (ASIC), 0 .. 15 within slot */
+    slotid_t	slot:4;		/* slot (Blade), 0 .. 15 within module */
+} geo_common_t;
 
 /* Additional fields for particular types of hardware */
 typedef struct geo_node_s {
-    geo_any_t	any;		/* No additional fields needed */
+    geo_common_t	common;		/* No additional fields needed */
 } geo_node_t;
 
 typedef struct geo_rtr_s {
-    geo_any_t	any;		/* No additional fields needed */
+    geo_common_t	common;		/* No additional fields needed */
 } geo_rtr_t;
 
 typedef struct geo_iocntl_s {
-    geo_any_t	any;		/* No additional fields needed */
+    geo_common_t	common;		/* No additional fields needed */
 } geo_iocntl_t;
 
 typedef struct geo_pcicard_s {
-    geo_iocntl_t	any;
+    geo_iocntl_t	common;
     char		bus;	/* Bus/widget number */
     char		slot;	/* PCI slot number */
 } geo_pcicard_t;
@@ -62,14 +64,14 @@ typedef struct geo_mem_s {
 
 
 typedef union geoid_u {
-    geo_any_t	any;
-    geo_node_t	node;
+    geo_common_t	common;
+    geo_node_t		node;
     geo_iocntl_t	iocntl;
     geo_pcicard_t	pcicard;
-    geo_rtr_t	rtr;
-    geo_cpu_t	cpu;
-    geo_mem_t	mem;
-    char	padsize[GEOID_SIZE];
+    geo_rtr_t		rtr;
+    geo_cpu_t		cpu;
+    geo_mem_t		mem;
+    char		padsize[GEOID_SIZE];
 } geoid_t;
 
 
@@ -104,19 +106,26 @@ typedef union geoid_u {
 #define INVALID_CNODEID         ((cnodeid_t)-1)
 #define INVALID_PNODEID         ((pnodeid_t)-1)
 #define INVALID_SLAB            (slabid_t)-1
+#define INVALID_SLOT            (slotid_t)-1
 #define INVALID_MODULE          ((moduleid_t)-1)
 #define INVALID_PARTID          ((partid_t)-1)
 
 static inline slabid_t geo_slab(geoid_t g)
 {
-	return (g.any.type == GEO_TYPE_INVALID) ?
-		INVALID_SLAB : g.any.slab;
+	return (g.common.type == GEO_TYPE_INVALID) ?
+		INVALID_SLAB : g.common.slab;
+}
+
+static inline slotid_t geo_slot(geoid_t g)
+{
+	return (g.common.type == GEO_TYPE_INVALID) ?
+		INVALID_SLOT : g.common.slot;
 }
 
 static inline moduleid_t geo_module(geoid_t g)
 {
-	return (g.any.type == GEO_TYPE_INVALID) ?
-		INVALID_MODULE : g.any.module;
+	return (g.common.type == GEO_TYPE_INVALID) ?
+		INVALID_MODULE : g.common.module;
 }
 
 extern geoid_t cnodeid_get_geoid(cnodeid_t cnode);
diff --git a/include/asm-ia64/sn/nodepda.h b/include/asm-ia64/sn/nodepda.h
index 2fbde33656e..13cc1002b29 100644
--- a/include/asm-ia64/sn/nodepda.h
+++ b/include/asm-ia64/sn/nodepda.h
@@ -3,7 +3,7 @@
  * License.  See the file "COPYING" in the main directory of this archive
  * for more details.
  *
- * Copyright (C) 1992 - 1997, 2000-2004 Silicon Graphics, Inc. All rights reserved.
+ * Copyright (C) 1992 - 1997, 2000-2005 Silicon Graphics, Inc. All rights reserved.
  */
 #ifndef _ASM_IA64_SN_NODEPDA_H
 #define _ASM_IA64_SN_NODEPDA_H
@@ -43,7 +43,7 @@ struct nodepda_s {
 	/*
 	 * The BTEs on this node are shared by the local cpus
 	 */
-	struct bteinfo_s	bte_if[BTES_PER_NODE];	/* Virtual Interface */
+	struct bteinfo_s	bte_if[MAX_BTES_PER_NODE];	/* Virtual Interface */
 	struct timer_list	bte_recovery_timer;
 	spinlock_t		bte_recovery_lock;
 
diff --git a/include/asm-ia64/sn/pcibus_provider_defs.h b/include/asm-ia64/sn/pcibus_provider_defs.h
new file mode 100644
index 00000000000..04e27d5b382
--- /dev/null
+++ b/include/asm-ia64/sn/pcibus_provider_defs.h
@@ -0,0 +1,52 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (C) 1992 - 1997, 2000-2004 Silicon Graphics, Inc. All rights reserved.
+ */
+#ifndef _ASM_IA64_SN_PCI_PCIBUS_PROVIDER_H
+#define _ASM_IA64_SN_PCI_PCIBUS_PROVIDER_H
+
+/*
+ * SN pci asic types.  Do not ever renumber these or reuse values.  The
+ * values must agree with what prom thinks they are.
+ */
+
+#define PCIIO_ASIC_TYPE_UNKNOWN	0
+#define PCIIO_ASIC_TYPE_PPB	1
+#define PCIIO_ASIC_TYPE_PIC	2
+#define PCIIO_ASIC_TYPE_TIOCP	3
+#define PCIIO_ASIC_TYPE_TIOCA	4
+
+#define PCIIO_ASIC_MAX_TYPES	5
+
+/*
+ * Common pciio bus provider data.  There should be one of these as the
+ * first field in any pciio based provider soft structure (e.g. pcibr_soft
+ * tioca_soft, etc).
+ */
+
+struct pcibus_bussoft {
+	uint32_t		bs_asic_type;	/* chipset type */
+	uint32_t		bs_xid;		/* xwidget id */
+	uint64_t		bs_persist_busnum; /* Persistent Bus Number */
+	uint64_t		bs_legacy_io;	/* legacy io pio addr */
+	uint64_t		bs_legacy_mem;	/* legacy mem pio addr */
+	uint64_t		bs_base;	/* widget base */
+	struct xwidget_info	*bs_xwidget_info;
+};
+
+/*
+ * SN pci bus indirection
+ */
+
+struct sn_pcibus_provider {
+	dma_addr_t	(*dma_map)(struct pci_dev *, unsigned long, size_t);
+	dma_addr_t	(*dma_map_consistent)(struct pci_dev *, unsigned long, size_t);
+	void		(*dma_unmap)(struct pci_dev *, dma_addr_t, int);
+	void *		(*bus_fixup)(struct pcibus_bussoft *);
+};
+
+extern struct sn_pcibus_provider *sn_pci_provider[];
+#endif				/* _ASM_IA64_SN_PCI_PCIBUS_PROVIDER_H */
diff --git a/include/asm-ia64/sn/pcidev.h b/include/asm-ia64/sn/pcidev.h
new file mode 100644
index 00000000000..ed4031d8081
--- /dev/null
+++ b/include/asm-ia64/sn/pcidev.h
@@ -0,0 +1,58 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (C) 1992 - 1997, 2000-2004 Silicon Graphics, Inc. All rights reserved.
+ */
+#ifndef _ASM_IA64_SN_PCI_PCIDEV_H
+#define _ASM_IA64_SN_PCI_PCIDEV_H
+
+#include <linux/pci.h>
+
+extern struct sn_irq_info **sn_irq;
+
+#define SN_PCIDEV_INFO(pci_dev) \
+        ((struct pcidev_info *)(pci_dev)->sysdata)
+
+/*
+ * Given a pci_bus, return the sn pcibus_bussoft struct.  Note that
+ * this only works for root busses, not for busses represented by PPB's.
+ */
+
+#define SN_PCIBUS_BUSSOFT(pci_bus) \
+        ((struct pcibus_bussoft *)(PCI_CONTROLLER((pci_bus))->platform_data))
+
+/*
+ * Given a struct pci_dev, return the sn pcibus_bussoft struct.  Note
+ * that this is not equivalent to SN_PCIBUS_BUSSOFT(pci_dev->bus) due
+ * due to possible PPB's in the path.
+ */
+
+#define SN_PCIDEV_BUSSOFT(pci_dev) \
+	(SN_PCIDEV_INFO(pci_dev)->pdi_host_pcidev_info->pdi_pcibus_info)
+
+#define SN_PCIDEV_BUSPROVIDER(pci_dev) \
+	(SN_PCIDEV_INFO(pci_dev)->pdi_provider)
+
+#define PCIIO_BUS_NONE	255      /* bus 255 reserved */
+#define PCIIO_SLOT_NONE 255
+#define PCIIO_FUNC_NONE 255
+#define PCIIO_VENDOR_ID_NONE	(-1)
+
+struct pcidev_info {
+	uint64_t		pdi_pio_mapped_addr[7]; /* 6 BARs PLUS 1 ROM */
+	uint64_t		pdi_slot_host_handle;	/* Bus and devfn Host pci_dev */
+
+	struct pcibus_bussoft	*pdi_pcibus_info;	/* Kernel common bus soft */
+	struct pcidev_info	*pdi_host_pcidev_info;	/* Kernel Host pci_dev */
+	struct pci_dev		*pdi_linux_pcidev;	/* Kernel pci_dev */
+
+	struct sn_irq_info	*pdi_sn_irq_info;
+	struct sn_pcibus_provider *pdi_provider;	/* sn pci ops */
+};
+
+extern void sn_irq_fixup(struct pci_dev *pci_dev,
+			 struct sn_irq_info *sn_irq_info);
+
+#endif				/* _ASM_IA64_SN_PCI_PCIDEV_H */
diff --git a/include/asm-ia64/sn/pda.h b/include/asm-ia64/sn/pda.h
index e940d3647c8..cd19f17bf91 100644
--- a/include/asm-ia64/sn/pda.h
+++ b/include/asm-ia64/sn/pda.h
@@ -3,7 +3,7 @@
  * License.  See the file "COPYING" in the main directory of this archive
  * for more details.
  *
- * Copyright (C) 1992 - 1997, 2000-2004 Silicon Graphics, Inc. All rights reserved.
+ * Copyright (C) 1992 - 1997, 2000-2005 Silicon Graphics, Inc. All rights reserved.
  */
 #ifndef _ASM_IA64_SN_PDA_H
 #define _ASM_IA64_SN_PDA_H
@@ -11,7 +11,6 @@
 #include <linux/cache.h>
 #include <asm/percpu.h>
 #include <asm/system.h>
-#include <asm/sn/bte.h>
 
 
 /*
diff --git a/include/asm-ia64/sn/shub_mmr.h b/include/asm-ia64/sn/shub_mmr.h
index 5c2fcf13d5c..2f885088e09 100644
--- a/include/asm-ia64/sn/shub_mmr.h
+++ b/include/asm-ia64/sn/shub_mmr.h
@@ -4,7 +4,7 @@
  * License.  See the file "COPYING" in the main directory of this archive
  * for more details.
  *
- * Copyright (c) 2001-2004 Silicon Graphics, Inc.  All rights reserved.
+ * Copyright (c) 2001-2005 Silicon Graphics, Inc.  All rights reserved.
  */
 
 #ifndef _ASM_IA64_SN_SHUB_MMR_H
@@ -129,6 +129,23 @@
 #define SH_EVENT_OCCURRED_II_INT1_SHFT           30
 #define SH_EVENT_OCCURRED_II_INT1_MASK           0x0000000040000000
 
+/*   SH2_EVENT_OCCURRED_EXTIO_INT2                                      */
+/*   Description:  Pending SHUB 2 EXT IO INT2                           */
+#define SH2_EVENT_OCCURRED_EXTIO_INT2_SHFT       33
+#define SH2_EVENT_OCCURRED_EXTIO_INT2_MASK       0x0000000200000000
+
+/*   SH2_EVENT_OCCURRED_EXTIO_INT3                                      */
+/*   Description:  Pending SHUB 2 EXT IO INT3                           */
+#define SH2_EVENT_OCCURRED_EXTIO_INT3_SHFT       34
+#define SH2_EVENT_OCCURRED_EXTIO_INT3_MASK       0x0000000400000000
+
+#define SH_ALL_INT_MASK \
+	(SH_EVENT_OCCURRED_UART_INT_MASK | SH_EVENT_OCCURRED_IPI_INT_MASK | \
+	 SH_EVENT_OCCURRED_II_INT0_MASK | SH_EVENT_OCCURRED_II_INT1_MASK | \
+	 SH_EVENT_OCCURRED_II_INT1_MASK | SH2_EVENT_OCCURRED_EXTIO_INT2_MASK | \
+	 SH2_EVENT_OCCURRED_EXTIO_INT3_MASK)
+
+
 /* ==================================================================== */
 /*                         LEDS                                         */
 /* ==================================================================== */
@@ -438,4 +455,22 @@
 #define SH_INT_CMPC		shubmmr(SH, INT_CMPC)
 #define SH_INT_CMPD		shubmmr(SH, INT_CMPD)
 
+/* ========================================================================== */
+/*                        Register "SH2_BT_ENG_CSR_0"                         */
+/*                    Engine 0 Control and Status Register                    */
+/* ========================================================================== */
+
+#define SH2_BT_ENG_CSR_0                         0x0000000030040000
+#define SH2_BT_ENG_SRC_ADDR_0                    0x0000000030040080
+#define SH2_BT_ENG_DEST_ADDR_0                   0x0000000030040100
+#define SH2_BT_ENG_NOTIF_ADDR_0                  0x0000000030040180
+
+/* ========================================================================== */
+/*                       BTE interfaces 1-3                                   */
+/* ========================================================================== */
+
+#define SH2_BT_ENG_CSR_1                         0x0000000030050000
+#define SH2_BT_ENG_CSR_2                         0x0000000030060000
+#define SH2_BT_ENG_CSR_3                         0x0000000030070000
+
 #endif /* _ASM_IA64_SN_SHUB_MMR_H */
diff --git a/include/asm-ia64/sn/sn_sal.h b/include/asm-ia64/sn/sn_sal.h
index 88c31b53dc0..f914f6da077 100644
--- a/include/asm-ia64/sn/sn_sal.h
+++ b/include/asm-ia64/sn/sn_sal.h
@@ -8,7 +8,7 @@
  * License.  See the file "COPYING" in the main directory of this archive
  * for more details.
  *
- * Copyright (c) 2000-2004 Silicon Graphics, Inc.  All rights reserved.
+ * Copyright (c) 2000-2005 Silicon Graphics, Inc.  All rights reserved.
  */
 
 
@@ -35,8 +35,8 @@
 #define  SN_SAL_PRINT_ERROR			   0x02000012
 #define  SN_SAL_SET_ERROR_HANDLING_FEATURES	   0x0200001a	// reentrant
 #define  SN_SAL_GET_FIT_COMPT			   0x0200001b	// reentrant
-#define  SN_SAL_GET_SN_INFO                        0x0200001c
 #define  SN_SAL_GET_SAPIC_INFO                     0x0200001d
+#define  SN_SAL_GET_SN_INFO                        0x0200001e
 #define  SN_SAL_CONSOLE_PUTC                       0x02000021
 #define  SN_SAL_CONSOLE_GETC                       0x02000022
 #define  SN_SAL_CONSOLE_PUTS                       0x02000023
@@ -64,6 +64,7 @@
 
 #define  SN_SAL_SYSCTL_IOBRICK_PCI_OP		   0x02000042	// reentrant
 #define	 SN_SAL_IROUTER_OP			   0x02000043
+#define  SN_SAL_SYSCTL_EVENT                       0x02000044
 #define  SN_SAL_IOIF_INTERRUPT			   0x0200004a
 #define  SN_SAL_HWPERF_OP			   0x02000050   // lock
 #define  SN_SAL_IOIF_ERROR_INTERRUPT		   0x02000051
@@ -76,7 +77,8 @@
 #define  SN_SAL_IOIF_GET_WIDGET_DMAFLUSH_LIST	   0x02000058
 
 #define SN_SAL_HUB_ERROR_INTERRUPT		   0x02000060
-
+#define SN_SAL_BTE_RECOVER			   0x02000061
+#define SN_SAL_IOIF_GET_PCI_TOPOLOGY	           0x02000062
 
 /*
  * Service-specific constants
@@ -849,6 +851,19 @@ ia64_sn_irtr_intr_disable(nasid_t nasid, int subch, u64 intr)
 	return (int) rv.v0;
 }
 
+/*
+ * Set up a node as the point of contact for system controller
+ * environmental event delivery.
+ */
+static inline int
+ia64_sn_sysctl_event_init(nasid_t nasid)
+{
+        struct ia64_sal_retval rv;
+        SAL_CALL_REENTRANT(rv, SN_SAL_SYSCTL_EVENT, (u64) nasid,
+			   0, 0, 0, 0, 0, 0);
+        return (int) rv.v0;
+}
+
 /**
  * ia64_sn_get_fit_compt - read a FIT entry from the PROM header
  * @nasid: NASID of node to read
@@ -1012,4 +1027,29 @@ ia64_sn_hwperf_op(nasid_t nasid, u64 opcode, u64 a0, u64 a1, u64 a2,
 	return (int) rv.status;
 }
 
+static inline int
+ia64_sn_ioif_get_pci_topology(u64 rack, u64 bay, u64 slot, u64 slab,
+			      u64 buf, u64 len)
+{
+	struct ia64_sal_retval rv;
+	SAL_CALL_NOLOCK(rv, SN_SAL_IOIF_GET_PCI_TOPOLOGY,
+		rack, bay, slot, slab, buf, len, 0);
+	return (int) rv.status;
+}
+
+/*
+ * BTE error recovery is implemented in SAL
+ */
+static inline int
+ia64_sn_bte_recovery(nasid_t nasid)
+{
+	struct ia64_sal_retval rv;
+
+	rv.status = 0;
+	SAL_CALL_NOLOCK(rv, SN_SAL_BTE_RECOVER, 0, 0, 0, 0, 0, 0, 0);
+	if (rv.status == SALRET_NOT_IMPLEMENTED)
+		return 0;
+	return (int) rv.status;
+}
+
 #endif /* _ASM_IA64_SN_SN_SAL_H */
diff --git a/include/asm-ia64/sn/tioca.h b/include/asm-ia64/sn/tioca.h
new file mode 100644
index 00000000000..bc1aacfb948
--- /dev/null
+++ b/include/asm-ia64/sn/tioca.h
@@ -0,0 +1,596 @@
+#ifndef _ASM_IA64_SN_TIO_TIOCA_H
+#define _ASM_IA64_SN_TIO_TIOCA_H
+
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (c) 2003-2005 Silicon Graphics, Inc. All rights reserved.
+ */
+
+
+#define TIOCA_PART_NUM	0xE020
+#define TIOCA_MFGR_NUM	0x24
+#define TIOCA_REV_A	0x1
+
+/*
+ * Register layout for TIO:CA.  See below for bitmasks for each register.
+ */
+
+struct tioca {
+	uint64_t	ca_id;				/* 0x000000 */
+	uint64_t	ca_control1;			/* 0x000008 */
+	uint64_t	ca_control2;			/* 0x000010 */
+	uint64_t	ca_status1;			/* 0x000018 */
+	uint64_t	ca_status2;			/* 0x000020 */
+	uint64_t	ca_gart_aperature;		/* 0x000028 */
+	uint64_t	ca_gfx_detach;			/* 0x000030 */
+	uint64_t	ca_inta_dest_addr;		/* 0x000038 */
+	uint64_t	ca_intb_dest_addr;		/* 0x000040 */
+	uint64_t	ca_err_int_dest_addr;		/* 0x000048 */
+	uint64_t	ca_int_status;			/* 0x000050 */
+	uint64_t	ca_int_status_alias;		/* 0x000058 */
+	uint64_t	ca_mult_error;			/* 0x000060 */
+	uint64_t	ca_mult_error_alias;		/* 0x000068 */
+	uint64_t	ca_first_error;			/* 0x000070 */
+	uint64_t	ca_int_mask;			/* 0x000078 */
+	uint64_t	ca_crm_pkterr_type;		/* 0x000080 */
+	uint64_t	ca_crm_pkterr_type_alias;	/* 0x000088 */
+	uint64_t	ca_crm_ct_error_detail_1;	/* 0x000090 */
+	uint64_t	ca_crm_ct_error_detail_2;	/* 0x000098 */
+	uint64_t	ca_crm_tnumto;			/* 0x0000A0 */
+	uint64_t	ca_gart_err;			/* 0x0000A8 */
+	uint64_t	ca_pcierr_type;			/* 0x0000B0 */
+	uint64_t	ca_pcierr_addr;			/* 0x0000B8 */
+
+	uint64_t	ca_pad_0000C0[3];		/* 0x0000{C0..D0} */
+
+	uint64_t	ca_pci_rd_buf_flush;		/* 0x0000D8 */
+	uint64_t	ca_pci_dma_addr_extn;		/* 0x0000E0 */
+	uint64_t	ca_agp_dma_addr_extn;		/* 0x0000E8 */
+	uint64_t	ca_force_inta;			/* 0x0000F0 */
+	uint64_t	ca_force_intb;			/* 0x0000F8 */
+	uint64_t	ca_debug_vector_sel;		/* 0x000100 */
+	uint64_t	ca_debug_mux_core_sel;		/* 0x000108 */
+	uint64_t	ca_debug_mux_pci_sel;		/* 0x000110 */
+	uint64_t	ca_debug_domain_sel;		/* 0x000118 */
+
+	uint64_t	ca_pad_000120[28];		/* 0x0001{20..F8} */
+
+	uint64_t	ca_gart_ptr_table;		/* 0x200 */
+	uint64_t	ca_gart_tlb_addr[8];		/* 0x2{08..40} */
+};
+
+/*
+ * Mask/shift definitions for TIO:CA registers.  The convention here is
+ * to mainly use the names as they appear in the "TIO AEGIS Programmers'
+ * Reference" with a CA_ prefix added.  Some exceptions were made to fix
+ * duplicate field names or to generalize fields that are common to
+ * different registers (ca_debug_mux_core_sel and ca_debug_mux_pci_sel for
+ * example).
+ *
+ * Fields consisting of a single bit have a single #define have a single
+ * macro declaration to mask the bit.  Fields consisting of multiple bits
+ * have two declarations: one to mask the proper bits in a register, and 
+ * a second with the suffix "_SHFT" to identify how far the mask needs to
+ * be shifted right to get its base value.
+ */
+
+/* ==== ca_control1 */
+#define CA_SYS_BIG_END			(1ull << 0)
+#define CA_DMA_AGP_SWAP			(1ull << 1)
+#define CA_DMA_PCI_SWAP			(1ull << 2)
+#define CA_PIO_IO_SWAP			(1ull << 3)
+#define CA_PIO_MEM_SWAP			(1ull << 4)
+#define CA_GFX_WR_SWAP			(1ull << 5)
+#define CA_AGP_FW_ENABLE		(1ull << 6)
+#define CA_AGP_CAL_CYCLE		(0x7ull << 7)
+#define CA_AGP_CAL_CYCLE_SHFT		7
+#define CA_AGP_CAL_PRSCL_BYP		(1ull << 10)
+#define CA_AGP_INIT_CAL_ENB		(1ull << 11)
+#define CA_INJ_ADDR_PERR		(1ull << 12)
+#define CA_INJ_DATA_PERR		(1ull << 13)
+	/* bits 15:14 unused */
+#define CA_PCIM_IO_NBE_AD		(0x7ull << 16)
+#define CA_PCIM_IO_NBE_AD_SHFT		16
+#define CA_PCIM_FAST_BTB_ENB		(1ull << 19)
+	/* bits 23:20 unused */
+#define CA_PIO_ADDR_OFFSET		(0xffull << 24)
+#define CA_PIO_ADDR_OFFSET_SHFT		24
+	/* bits 35:32 unused */
+#define CA_AGPDMA_OP_COMBDELAY		(0x1full << 36)
+#define CA_AGPDMA_OP_COMBDELAY_SHFT	36
+	/* bit 41 unused */
+#define CA_AGPDMA_OP_ENB_COMBDELAY	(1ull << 42)
+#define	CA_PCI_INT_LPCNT		(0xffull << 44)
+#define CA_PCI_INT_LPCNT_SHFT		44
+	/* bits 63:52 unused */
+
+/* ==== ca_control2 */
+#define CA_AGP_LATENCY_TO		(0xffull << 0)
+#define CA_AGP_LATENCY_TO_SHFT		0
+#define CA_PCI_LATENCY_TO		(0xffull << 8)
+#define CA_PCI_LATENCY_TO_SHFT		8
+#define CA_PCI_MAX_RETRY		(0x3ffull << 16)
+#define CA_PCI_MAX_RETRY_SHFT		16
+	/* bits 27:26 unused */
+#define CA_RT_INT_EN			(0x3ull << 28)
+#define CA_RT_INT_EN_SHFT			28
+#define CA_MSI_INT_ENB			(1ull << 30)
+#define CA_PCI_ARB_ERR_ENB		(1ull << 31)
+#define CA_GART_MEM_PARAM		(0x3ull << 32)
+#define CA_GART_MEM_PARAM_SHFT		32
+#define CA_GART_RD_PREFETCH_ENB		(1ull << 34)
+#define CA_GART_WR_PREFETCH_ENB		(1ull << 35)
+#define CA_GART_FLUSH_TLB		(1ull << 36)
+	/* bits 39:37 unused */
+#define CA_CRM_TNUMTO_PERIOD		(0x1fffull << 40)
+#define CA_CRM_TNUMTO_PERIOD_SHFT	40
+	/* bits 55:53 unused */
+#define CA_CRM_TNUMTO_ENB		(1ull << 56)
+#define CA_CRM_PRESCALER_BYP		(1ull << 57)
+	/* bits 59:58 unused */
+#define CA_CRM_MAX_CREDIT		(0x7ull << 60)
+#define CA_CRM_MAX_CREDIT_SHFT		60
+	/* bit 63 unused */
+
+/* ==== ca_status1 */
+#define CA_CORELET_ID			(0x3ull << 0)
+#define CA_CORELET_ID_SHFT		0
+#define CA_INTA_N			(1ull << 2)
+#define CA_INTB_N			(1ull << 3)
+#define CA_CRM_CREDIT_AVAIL		(0x7ull << 4)
+#define CA_CRM_CREDIT_AVAIL_SHFT	4
+	/* bit 7 unused */
+#define CA_CRM_SPACE_AVAIL		(0x7full << 8)
+#define CA_CRM_SPACE_AVAIL_SHFT		8
+	/* bit 15 unused */
+#define CA_GART_TLB_VAL			(0xffull << 16)
+#define CA_GART_TLB_VAL_SHFT		16
+	/* bits 63:24 unused */
+
+/* ==== ca_status2 */
+#define CA_GFX_CREDIT_AVAIL		(0xffull << 0)
+#define CA_GFX_CREDIT_AVAIL_SHFT	0
+#define CA_GFX_OPQ_AVAIL		(0xffull << 8)
+#define CA_GFX_OPQ_AVAIL_SHFT		8
+#define CA_GFX_WRBUFF_AVAIL		(0xffull << 16)
+#define CA_GFX_WRBUFF_AVAIL_SHFT	16
+#define CA_ADMA_OPQ_AVAIL		(0xffull << 24)
+#define CA_ADMA_OPQ_AVAIL_SHFT		24
+#define CA_ADMA_WRBUFF_AVAIL		(0xffull << 32)
+#define CA_ADMA_WRBUFF_AVAIL_SHFT	32
+#define CA_ADMA_RDBUFF_AVAIL		(0x7full << 40)
+#define CA_ADMA_RDBUFF_AVAIL_SHFT	40
+#define CA_PCI_PIO_OP_STAT		(1ull << 47)
+#define CA_PDMA_OPQ_AVAIL		(0xfull << 48)
+#define CA_PDMA_OPQ_AVAIL_SHFT		48
+#define CA_PDMA_WRBUFF_AVAIL		(0xfull << 52)
+#define CA_PDMA_WRBUFF_AVAIL_SHFT	52
+#define CA_PDMA_RDBUFF_AVAIL		(0x3ull << 56)
+#define CA_PDMA_RDBUFF_AVAIL_SHFT	56
+	/* bits 63:58 unused */
+
+/* ==== ca_gart_aperature */
+#define CA_GART_AP_ENB_AGP		(1ull << 0)
+#define CA_GART_PAGE_SIZE		(1ull << 1)
+#define CA_GART_AP_ENB_PCI		(1ull << 2)
+	/* bits 11:3 unused */
+#define CA_GART_AP_SIZE			(0x3ffull << 12)
+#define CA_GART_AP_SIZE_SHFT		12
+#define CA_GART_AP_BASE			(0x3ffffffffffull << 22)
+#define CA_GART_AP_BASE_SHFT		22
+
+/* ==== ca_inta_dest_addr
+   ==== ca_intb_dest_addr 
+   ==== ca_err_int_dest_addr */
+	/* bits 2:0 unused */
+#define CA_INT_DEST_ADDR		(0x7ffffffffffffull << 3)
+#define CA_INT_DEST_ADDR_SHFT		3
+	/* bits 55:54 unused */
+#define CA_INT_DEST_VECT		(0xffull << 56)
+#define CA_INT_DEST_VECT_SHFT		56
+
+/* ==== ca_int_status */
+/* ==== ca_int_status_alias */
+/* ==== ca_mult_error */
+/* ==== ca_mult_error_alias */
+/* ==== ca_first_error */
+/* ==== ca_int_mask */
+#define CA_PCI_ERR			(1ull << 0)
+	/* bits 3:1 unused */
+#define CA_GART_FETCH_ERR		(1ull << 4)
+#define CA_GFX_WR_OVFLW			(1ull << 5)
+#define CA_PIO_REQ_OVFLW		(1ull << 6)
+#define CA_CRM_PKTERR			(1ull << 7)
+#define CA_CRM_DVERR			(1ull << 8)
+#define CA_TNUMTO			(1ull << 9)
+#define CA_CXM_RSP_CRED_OVFLW		(1ull << 10)
+#define CA_CXM_REQ_CRED_OVFLW		(1ull << 11)
+#define CA_PIO_INVALID_ADDR		(1ull << 12)
+#define CA_PCI_ARB_TO			(1ull << 13)
+#define CA_AGP_REQ_OFLOW		(1ull << 14)
+#define CA_SBA_TYPE1_ERR		(1ull << 15)
+	/* bit 16 unused */
+#define CA_INTA				(1ull << 17)
+#define CA_INTB				(1ull << 18)
+#define CA_MULT_INTA			(1ull << 19)
+#define CA_MULT_INTB			(1ull << 20)
+#define CA_GFX_CREDIT_OVFLW		(1ull << 21)
+	/* bits 63:22 unused */
+
+/* ==== ca_crm_pkterr_type */
+/* ==== ca_crm_pkterr_type_alias */
+#define CA_CRM_PKTERR_SBERR_HDR		(1ull << 0)
+#define CA_CRM_PKTERR_DIDN		(1ull << 1)
+#define CA_CRM_PKTERR_PACTYPE		(1ull << 2)
+#define CA_CRM_PKTERR_INV_TNUM		(1ull << 3)
+#define CA_CRM_PKTERR_ADDR_RNG		(1ull << 4)
+#define CA_CRM_PKTERR_ADDR_ALGN		(1ull << 5)
+#define CA_CRM_PKTERR_HDR_PARAM		(1ull << 6)
+#define CA_CRM_PKTERR_CW_ERR		(1ull << 7)
+#define CA_CRM_PKTERR_SBERR_NH		(1ull << 8)
+#define CA_CRM_PKTERR_EARLY_TERM	(1ull << 9)
+#define CA_CRM_PKTERR_EARLY_TAIL	(1ull << 10)
+#define CA_CRM_PKTERR_MSSNG_TAIL	(1ull << 11)
+#define CA_CRM_PKTERR_MSSNG_HDR		(1ull << 12)
+	/* bits 15:13 unused */
+#define CA_FIRST_CRM_PKTERR_SBERR_HDR	(1ull << 16)
+#define CA_FIRST_CRM_PKTERR_DIDN	(1ull << 17)
+#define CA_FIRST_CRM_PKTERR_PACTYPE	(1ull << 18)
+#define CA_FIRST_CRM_PKTERR_INV_TNUM	(1ull << 19)
+#define CA_FIRST_CRM_PKTERR_ADDR_RNG	(1ull << 20)
+#define CA_FIRST_CRM_PKTERR_ADDR_ALGN	(1ull << 21)
+#define CA_FIRST_CRM_PKTERR_HDR_PARAM	(1ull << 22)
+#define CA_FIRST_CRM_PKTERR_CW_ERR	(1ull << 23)
+#define CA_FIRST_CRM_PKTERR_SBERR_NH	(1ull << 24)
+#define CA_FIRST_CRM_PKTERR_EARLY_TERM	(1ull << 25)
+#define CA_FIRST_CRM_PKTERR_EARLY_TAIL	(1ull << 26)
+#define CA_FIRST_CRM_PKTERR_MSSNG_TAIL	(1ull << 27)
+#define CA_FIRST_CRM_PKTERR_MSSNG_HDR	(1ull << 28)
+	/* bits 63:29 unused */
+
+/* ==== ca_crm_ct_error_detail_1 */
+#define CA_PKT_TYPE			(0xfull << 0)
+#define CA_PKT_TYPE_SHFT		0
+#define CA_SRC_ID			(0x3ull << 4)
+#define CA_SRC_ID_SHFT			4
+#define CA_DATA_SZ			(0x3ull << 6)
+#define CA_DATA_SZ_SHFT			6
+#define CA_TNUM				(0xffull << 8)
+#define CA_TNUM_SHFT			8
+#define CA_DW_DATA_EN			(0xffull << 16)
+#define CA_DW_DATA_EN_SHFT		16
+#define CA_GFX_CRED			(0xffull << 24)
+#define CA_GFX_CRED_SHFT		24
+#define CA_MEM_RD_PARAM			(0x3ull << 32)
+#define CA_MEM_RD_PARAM_SHFT		32
+#define CA_PIO_OP			(1ull << 34)
+#define CA_CW_ERR			(1ull << 35)
+	/* bits 62:36 unused */
+#define CA_VALID			(1ull << 63)
+
+/* ==== ca_crm_ct_error_detail_2 */
+	/* bits 2:0 unused */
+#define CA_PKT_ADDR			(0x1fffffffffffffull << 3)
+#define CA_PKT_ADDR_SHFT		3
+	/* bits 63:56 unused */
+
+/* ==== ca_crm_tnumto */
+#define CA_CRM_TNUMTO_VAL		(0xffull << 0)
+#define CA_CRM_TNUMTO_VAL_SHFT		0
+#define CA_CRM_TNUMTO_WR		(1ull << 8)
+	/* bits 63:9 unused */
+
+/* ==== ca_gart_err */
+#define CA_GART_ERR_SOURCE		(0x3ull << 0)
+#define CA_GART_ERR_SOURCE_SHFT		0
+	/* bits 3:2 unused */
+#define CA_GART_ERR_ADDR		(0xfffffffffull << 4)
+#define CA_GART_ERR_ADDR_SHFT		4
+	/* bits 63:40 unused */
+
+/* ==== ca_pcierr_type */
+#define CA_PCIERR_DATA			(0xffffffffull << 0)
+#define CA_PCIERR_DATA_SHFT		0
+#define CA_PCIERR_ENB			(0xfull << 32)
+#define CA_PCIERR_ENB_SHFT		32
+#define CA_PCIERR_CMD			(0xfull << 36)
+#define CA_PCIERR_CMD_SHFT		36
+#define CA_PCIERR_A64			(1ull << 40)
+#define CA_PCIERR_SLV_SERR		(1ull << 41)
+#define CA_PCIERR_SLV_WR_PERR		(1ull << 42)
+#define CA_PCIERR_SLV_RD_PERR		(1ull << 43)
+#define CA_PCIERR_MST_SERR		(1ull << 44)
+#define CA_PCIERR_MST_WR_PERR		(1ull << 45)
+#define CA_PCIERR_MST_RD_PERR		(1ull << 46)
+#define CA_PCIERR_MST_MABT		(1ull << 47)
+#define CA_PCIERR_MST_TABT		(1ull << 48)
+#define CA_PCIERR_MST_RETRY_TOUT	(1ull << 49)
+
+#define CA_PCIERR_TYPES \
+	(CA_PCIERR_A64|CA_PCIERR_SLV_SERR| \
+	 CA_PCIERR_SLV_WR_PERR|CA_PCIERR_SLV_RD_PERR| \
+	 CA_PCIERR_MST_SERR|CA_PCIERR_MST_WR_PERR|CA_PCIERR_MST_RD_PERR| \
+	 CA_PCIERR_MST_MABT|CA_PCIERR_MST_TABT|CA_PCIERR_MST_RETRY_TOUT)
+
+	/* bits 63:50 unused */
+
+/* ==== ca_pci_dma_addr_extn */
+#define CA_UPPER_NODE_OFFSET		(0x3full << 0)
+#define CA_UPPER_NODE_OFFSET_SHFT	0
+	/* bits 7:6 unused */
+#define CA_CHIPLET_ID			(0x3ull << 8)
+#define CA_CHIPLET_ID_SHFT		8
+	/* bits 11:10 unused */
+#define CA_PCI_DMA_NODE_ID		(0xffffull << 12)
+#define CA_PCI_DMA_NODE_ID_SHFT		12
+	/* bits 27:26 unused */
+#define CA_PCI_DMA_PIO_MEM_TYPE		(1ull << 28)
+	/* bits 63:29 unused */
+
+
+/* ==== ca_agp_dma_addr_extn */
+	/* bits 19:0 unused */
+#define CA_AGP_DMA_NODE_ID		(0xffffull << 20)
+#define CA_AGP_DMA_NODE_ID_SHFT		20
+	/* bits 27:26 unused */
+#define CA_AGP_DMA_PIO_MEM_TYPE		(1ull << 28)
+	/* bits 63:29 unused */
+
+/* ==== ca_debug_vector_sel */
+#define CA_DEBUG_MN_VSEL		(0xfull << 0)
+#define CA_DEBUG_MN_VSEL_SHFT		0
+#define CA_DEBUG_PP_VSEL		(0xfull << 4)
+#define CA_DEBUG_PP_VSEL_SHFT		4
+#define CA_DEBUG_GW_VSEL		(0xfull << 8)
+#define CA_DEBUG_GW_VSEL_SHFT		8
+#define CA_DEBUG_GT_VSEL		(0xfull << 12)
+#define CA_DEBUG_GT_VSEL_SHFT		12
+#define CA_DEBUG_PD_VSEL		(0xfull << 16)
+#define CA_DEBUG_PD_VSEL_SHFT		16
+#define CA_DEBUG_AD_VSEL		(0xfull << 20)
+#define CA_DEBUG_AD_VSEL_SHFT		20
+#define CA_DEBUG_CX_VSEL		(0xfull << 24)
+#define CA_DEBUG_CX_VSEL_SHFT		24
+#define CA_DEBUG_CR_VSEL		(0xfull << 28)
+#define CA_DEBUG_CR_VSEL_SHFT		28
+#define CA_DEBUG_BA_VSEL		(0xfull << 32)
+#define CA_DEBUG_BA_VSEL_SHFT		32
+#define CA_DEBUG_PE_VSEL		(0xfull << 36)
+#define CA_DEBUG_PE_VSEL_SHFT		36
+#define CA_DEBUG_BO_VSEL		(0xfull << 40)
+#define CA_DEBUG_BO_VSEL_SHFT		40
+#define CA_DEBUG_BI_VSEL		(0xfull << 44)
+#define CA_DEBUG_BI_VSEL_SHFT		44
+#define CA_DEBUG_AS_VSEL		(0xfull << 48)
+#define CA_DEBUG_AS_VSEL_SHFT		48
+#define CA_DEBUG_PS_VSEL		(0xfull << 52)
+#define CA_DEBUG_PS_VSEL_SHFT		52
+#define CA_DEBUG_PM_VSEL		(0xfull << 56)
+#define CA_DEBUG_PM_VSEL_SHFT		56
+	/* bits 63:60 unused */
+
+/* ==== ca_debug_mux_core_sel */
+/* ==== ca_debug_mux_pci_sel */
+#define CA_DEBUG_MSEL0			(0x7ull << 0)
+#define CA_DEBUG_MSEL0_SHFT		0
+	/* bit 3 unused */
+#define CA_DEBUG_NSEL0			(0x7ull << 4)
+#define CA_DEBUG_NSEL0_SHFT		4
+	/* bit 7 unused */
+#define CA_DEBUG_MSEL1			(0x7ull << 8)
+#define CA_DEBUG_MSEL1_SHFT		8
+	/* bit 11 unused */
+#define CA_DEBUG_NSEL1			(0x7ull << 12)
+#define CA_DEBUG_NSEL1_SHFT		12
+	/* bit 15 unused */
+#define CA_DEBUG_MSEL2			(0x7ull << 16)
+#define CA_DEBUG_MSEL2_SHFT		16
+	/* bit 19 unused */
+#define CA_DEBUG_NSEL2			(0x7ull << 20)
+#define CA_DEBUG_NSEL2_SHFT		20
+	/* bit 23 unused */
+#define CA_DEBUG_MSEL3			(0x7ull << 24)
+#define CA_DEBUG_MSEL3_SHFT		24
+	/* bit 27 unused */
+#define CA_DEBUG_NSEL3			(0x7ull << 28)
+#define CA_DEBUG_NSEL3_SHFT		28
+	/* bit 31 unused */
+#define CA_DEBUG_MSEL4			(0x7ull << 32)
+#define CA_DEBUG_MSEL4_SHFT		32
+	/* bit 35 unused */
+#define CA_DEBUG_NSEL4			(0x7ull << 36)
+#define CA_DEBUG_NSEL4_SHFT		36
+	/* bit 39 unused */
+#define CA_DEBUG_MSEL5			(0x7ull << 40)
+#define CA_DEBUG_MSEL5_SHFT		40
+	/* bit 43 unused */
+#define CA_DEBUG_NSEL5			(0x7ull << 44)
+#define CA_DEBUG_NSEL5_SHFT		44
+	/* bit 47 unused */
+#define CA_DEBUG_MSEL6			(0x7ull << 48)
+#define CA_DEBUG_MSEL6_SHFT		48
+	/* bit 51 unused */
+#define CA_DEBUG_NSEL6			(0x7ull << 52)
+#define CA_DEBUG_NSEL6_SHFT		52
+	/* bit 55 unused */
+#define CA_DEBUG_MSEL7			(0x7ull << 56)
+#define CA_DEBUG_MSEL7_SHFT		56
+	/* bit 59 unused */
+#define CA_DEBUG_NSEL7			(0x7ull << 60)
+#define CA_DEBUG_NSEL7_SHFT		60
+	/* bit 63 unused */
+
+
+/* ==== ca_debug_domain_sel */
+#define CA_DEBUG_DOMAIN_L		(1ull << 0)
+#define CA_DEBUG_DOMAIN_H		(1ull << 1)
+	/* bits 63:2 unused */
+
+/* ==== ca_gart_ptr_table */
+#define CA_GART_PTR_VAL			(1ull << 0)
+	/* bits 11:1 unused */
+#define CA_GART_PTR_ADDR		(0xfffffffffffull << 12)
+#define CA_GART_PTR_ADDR_SHFT		12
+	/* bits 63:56 unused */
+
+/* ==== ca_gart_tlb_addr[0-7] */
+#define CA_GART_TLB_ADDR		(0xffffffffffffffull << 0)
+#define CA_GART_TLB_ADDR_SHFT		0
+	/* bits 62:56 unused */
+#define CA_GART_TLB_ENTRY_VAL		(1ull << 63)
+
+/*
+ * PIO address space ranges for TIO:CA
+ */
+
+/* CA internal registers */
+#define CA_PIO_ADMIN			0x00000000
+#define CA_PIO_ADMIN_LEN		0x00010000
+
+/* GFX Write Buffer - Diagnostics */
+#define CA_PIO_GFX			0x00010000
+#define CA_PIO_GFX_LEN			0x00010000
+
+/* AGP DMA Write Buffer - Diagnostics */
+#define CA_PIO_AGP_DMAWRITE		0x00020000
+#define CA_PIO_AGP_DMAWRITE_LEN		0x00010000
+
+/* AGP DMA READ Buffer - Diagnostics */
+#define CA_PIO_AGP_DMAREAD		0x00030000
+#define CA_PIO_AGP_DMAREAD_LEN		0x00010000
+
+/* PCI Config Type 0 */
+#define CA_PIO_PCI_TYPE0_CONFIG		0x01000000
+#define CA_PIO_PCI_TYPE0_CONFIG_LEN	0x01000000
+
+/* PCI Config Type 1 */
+#define CA_PIO_PCI_TYPE1_CONFIG		0x02000000
+#define CA_PIO_PCI_TYPE1_CONFIG_LEN	0x01000000
+
+/* PCI I/O Cycles - mapped to PCI Address 0x00000000-0x04ffffff */
+#define CA_PIO_PCI_IO			0x03000000
+#define CA_PIO_PCI_IO_LEN		0x05000000
+
+/* PCI MEM Cycles - mapped to PCI with CA_PIO_ADDR_OFFSET of ca_control1 */
+/*	use Fast Write if enabled and coretalk packet type is a GFX request */
+#define CA_PIO_PCI_MEM_OFFSET		0x08000000
+#define CA_PIO_PCI_MEM_OFFSET_LEN	0x08000000
+
+/* PCI MEM Cycles - mapped to PCI Address 0x00000000-0xbfffffff */
+/*	use Fast Write if enabled and coretalk packet type is a GFX request */
+#define CA_PIO_PCI_MEM			0x40000000
+#define CA_PIO_PCI_MEM_LEN		0xc0000000
+
+/*
+ * DMA space
+ *
+ * The CA aperature (ie. bus address range) mapped by the GART is segmented into
+ * two parts.  The lower portion of the aperature is used for mapping 32 bit
+ * PCI addresses which are managed by the dma interfaces in this file.  The
+ * upper poprtion of the aperature is used for mapping 48 bit AGP addresses.
+ * The AGP portion of the aperature is managed by the agpgart_be.c driver
+ * in drivers/linux/agp.  There are ca-specific hooks in that driver to
+ * manipulate the gart, but management of the AGP portion of the aperature
+ * is the responsibility of that driver.
+ *
+ * CA allows three main types of DMA mapping:
+ *
+ * PCI 64-bit	Managed by this driver
+ * PCI 32-bit 	Managed by this driver
+ * AGP 48-bit	Managed by hooks in the /dev/agpgart driver
+ *
+ * All of the above can optionally be remapped through the GART.  The following
+ * table lists the combinations of addressing types and GART remapping that
+ * is currently supported by the driver (h/w supports all, s/w limits this):
+ *
+ *		PCI64		PCI32		AGP48
+ * GART		no		yes		yes
+ * Direct	yes		yes		no
+ *
+ * GART remapping of PCI64 is not done because there is no need to.  The
+ * 64 bit PCI address holds all of the information necessary to target any
+ * memory in the system.
+ *
+ * AGP48 is always mapped through the GART.  Management of the AGP48 portion
+ * of the aperature is the responsibility of code in the agpgart_be driver.
+ *
+ * The non-64 bit bus address space will currently be partitioned like this:
+ *
+ *	0xffff_ffff_ffff	+--------
+ *				| AGP48 direct
+ *				| Space managed by this driver
+ *	CA_AGP_DIRECT_BASE	+--------
+ *				| AGP GART mapped (gfx aperature)
+ *				| Space managed by /dev/agpgart driver
+ *				| This range is exposed to the agpgart
+ * 				| driver as the "graphics aperature"
+ *	CA_AGP_MAPPED_BASE	+-----
+ *				| PCI GART mapped
+ *				| Space managed by this driver		
+ *	CA_PCI32_MAPPED_BASE	+----
+ *				| PCI32 direct
+ *				| Space managed by this driver
+ *	0xC000_0000		+--------
+ *	(CA_PCI32_DIRECT_BASE)
+ *
+ * The bus address range CA_PCI32_MAPPED_BASE through CA_AGP_DIRECT_BASE
+ * is what we call the CA aperature.  Addresses falling in this range will
+ * be remapped using the GART.
+ *
+ * The bus address range CA_AGP_MAPPED_BASE through CA_AGP_DIRECT_BASE
+ * is what we call the graphics aperature.  This is a subset of the CA
+ * aperature and is under the control of the agpgart_be driver.
+ *
+ * CA_PCI32_MAPPED_BASE, CA_AGP_MAPPED_BASE, and CA_AGP_DIRECT_BASE are
+ * somewhat arbitrary values.  The known constraints on choosing these is:
+ *
+ * 1)  CA_AGP_DIRECT_BASE-CA_PCI32_MAPPED_BASE+1 (the CA aperature size)
+ *     must be one of the values supported by the ca_gart_aperature register.
+ *     Currently valid values are: 4MB through 4096MB in powers of 2 increments
+ *
+ * 2)  CA_AGP_DIRECT_BASE-CA_AGP_MAPPED_BASE+1 (the gfx aperature size)
+ *     must be in MB units since that's what the agpgart driver assumes.
+ */
+
+/*
+ * Define Bus DMA ranges.  These are configurable (see constraints above)
+ * and will probably need tuning based on experience.
+ */
+
+
+/*
+ * 11/24/03
+ * CA has an addressing glitch w.r.t. PCI direct 32 bit DMA that makes it
+ * generally unusable.  The problem is that for PCI direct 32 
+ * DMA's, all 32 bits of the bus address are used to form the lower 32 bits
+ * of the coretalk address, and coretalk bits 38:32 come from a register.
+ * Since only PCI bus addresses 0xC0000000-0xFFFFFFFF (1GB) are available
+ * for DMA (the rest is allocated to PIO), host node addresses need to be
+ * such that their lower 32 bits fall in the 0xC0000000-0xffffffff range
+ * as well.  So there can be no PCI32 direct DMA below 3GB!!  For this
+ * reason we set the CA_PCI32_DIRECT_SIZE to 0 which essentially makes
+ * tioca_dma_direct32() a noop but preserves the code flow should this issue
+ * be fixed in a respin.
+ *
+ * For now, all PCI32 DMA's must be mapped through the GART.
+ */
+
+#define CA_PCI32_DIRECT_BASE	0xC0000000UL	/* BASE not configurable */
+#define CA_PCI32_DIRECT_SIZE	0x00000000UL	/* 0 MB */
+
+#define CA_PCI32_MAPPED_BASE	0xC0000000UL
+#define CA_PCI32_MAPPED_SIZE	0x40000000UL	/* 2GB */
+
+#define CA_AGP_MAPPED_BASE	0x80000000UL
+#define CA_AGP_MAPPED_SIZE	0x40000000UL	/* 2GB */
+
+#define CA_AGP_DIRECT_BASE	0x40000000UL	/* 2GB */
+#define CA_AGP_DIRECT_SIZE	0x40000000UL
+
+#define CA_APERATURE_BASE	(CA_AGP_MAPPED_BASE)
+#define CA_APERATURE_SIZE	(CA_AGP_MAPPED_SIZE+CA_PCI32_MAPPED_SIZE)
+
+#endif  /* _ASM_IA64_SN_TIO_TIOCA_H */
diff --git a/include/asm-ia64/sn/tioca_provider.h b/include/asm-ia64/sn/tioca_provider.h
new file mode 100644
index 00000000000..b6acc22ab23
--- /dev/null
+++ b/include/asm-ia64/sn/tioca_provider.h
@@ -0,0 +1,206 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (c) 2003-2005 Silicon Graphics, Inc. All rights reserved.
+ */
+
+#ifndef _ASM_IA64_SN_TIO_CA_AGP_PROVIDER_H
+#define _ASM_IA64_SN_TIO_CA_AGP_PROVIDER_H
+
+#include <asm/sn/tioca.h>
+
+/*
+ * WAR enables
+ * Defines for individual WARs. Each is a bitmask of applicable
+ * part revision numbers. (1 << 1) == rev A, (1 << 2) == rev B,
+ * (3 << 1) == (rev A or rev B), etc
+ */
+
+#define TIOCA_WAR_ENABLED(pv, tioca_common) \
+	((1 << tioca_common->ca_rev) & pv)
+
+  /* TIO:ICE:FRZ:Freezer loses a PIO data ucred on PIO RD RSP with CW error */
+#define PV907908 (1 << 1)
+  /* ATI config space problems after BIOS execution starts */
+#define PV908234 (1 << 1)
+  /* CA:AGPDMA write request data mismatch with ABC1CL merge */
+#define PV895469 (1 << 1)
+  /* TIO:CA TLB invalidate of written GART entries possibly not occuring in CA*/
+#define PV910244 (1 << 1)
+
+struct tioca_dmamap{
+	struct list_head	cad_list;	/* headed by ca_list */
+
+	dma_addr_t		cad_dma_addr;	/* Linux dma handle */
+	uint			cad_gart_entry; /* start entry in ca_gart_pagemap */
+	uint			cad_gart_size;	/* #entries for this map */
+};
+
+/*
+ * Kernel only fields.  Prom may look at this stuff for debugging only.
+ * Access this structure through the ca_kernel_private ptr.
+ */
+
+struct tioca_common ;
+
+struct tioca_kernel {
+	struct tioca_common	*ca_common;	/* tioca this belongs to */
+	struct list_head	ca_list;	/* list of all ca's */
+	struct list_head	ca_dmamaps;
+	spinlock_t		ca_lock;	/* Kernel lock */
+	cnodeid_t		ca_closest_node;
+	struct list_head	*ca_devices;	/* bus->devices */
+
+	/*
+	 * General GART stuff
+	 */
+	uint64_t	ca_ap_size;		/* size of aperature in bytes */
+	uint32_t	ca_gart_entries;	/* # uint64_t entries in gart */
+	uint32_t	ca_ap_pagesize; 	/* aperature page size in bytes */
+	uint64_t	ca_ap_bus_base; 	/* bus address of CA aperature */
+	uint64_t	ca_gart_size;		/* gart size in bytes */
+	uint64_t	*ca_gart;		/* gart table vaddr */
+	uint64_t	ca_gart_coretalk_addr;	/* gart coretalk addr */
+	uint8_t		ca_gart_iscoherent;	/* used in tioca_tlbflush */
+
+	/* PCI GART convenience values */
+	uint64_t	ca_pciap_base;		/* pci aperature bus base address */
+	uint64_t	ca_pciap_size;		/* pci aperature size (bytes) */
+	uint64_t	ca_pcigart_base;	/* gfx GART bus base address */
+	uint64_t	*ca_pcigart;		/* gfx GART vm address */
+	uint32_t	ca_pcigart_entries;
+	uint32_t	ca_pcigart_start;	/* PCI start index in ca_gart */
+	void		*ca_pcigart_pagemap;
+
+	/* AGP GART convenience values */
+	uint64_t	ca_gfxap_base;		/* gfx aperature bus base address */
+	uint64_t	ca_gfxap_size;		/* gfx aperature size (bytes) */
+	uint64_t	ca_gfxgart_base;	/* gfx GART bus base address */
+	uint64_t	*ca_gfxgart;		/* gfx GART vm address */
+	uint32_t	ca_gfxgart_entries;
+	uint32_t	ca_gfxgart_start;	/* agpgart start index in ca_gart */
+};
+
+/*
+ * Common tioca info shared between kernel and prom
+ *
+ * DO NOT CHANGE THIS STRUCT WITHOUT MAKING CORRESPONDING CHANGES
+ * TO THE PROM VERSION.
+ */
+
+struct tioca_common {
+	struct pcibus_bussoft	ca_common;	/* common pciio header */
+
+	uint32_t		ca_rev;
+	uint32_t		ca_closest_nasid;
+
+	uint64_t		ca_prom_private;
+	uint64_t		ca_kernel_private;
+};
+
+/**
+ * tioca_paddr_to_gart - Convert an SGI coretalk address to a CA GART entry
+ * @paddr: page address to convert
+ *
+ * Convert a system [coretalk] address to a GART entry.  GART entries are
+ * formed using the following:
+ *
+ *     data = ( (1<<63) |  ( (REMAP_NODE_ID << 40) | (MD_CHIPLET_ID << 38) | 
+ * (REMAP_SYS_ADDR) ) >> 12 )
+ *
+ * DATA written to 1 GART TABLE Entry in system memory is remapped system
+ * addr for 1 page 
+ *
+ * The data is for coretalk address format right shifted 12 bits with a
+ * valid bit.
+ *
+ *	GART_TABLE_ENTRY [ 25:0 ]  -- REMAP_SYS_ADDRESS[37:12].
+ *	GART_TABLE_ENTRY [ 27:26 ] -- SHUB MD chiplet id.
+ *	GART_TABLE_ENTRY [ 41:28 ] -- REMAP_NODE_ID.
+ *	GART_TABLE_ENTRY [ 63 ]    -- Valid Bit 
+ */
+static inline u64
+tioca_paddr_to_gart(unsigned long paddr)
+{
+	/*
+	 * We are assuming right now that paddr already has the correct
+	 * format since the address from xtalk_dmaXXX should already have
+	 * NODE_ID, CHIPLET_ID, and SYS_ADDR in the correct locations.
+	 */
+
+	return ((paddr) >> 12) | (1UL << 63);
+}
+
+/**
+ * tioca_physpage_to_gart - Map a host physical page for SGI CA based DMA
+ * @page_addr: system page address to map
+ */
+
+static inline unsigned long
+tioca_physpage_to_gart(uint64_t page_addr)
+{
+	uint64_t coretalk_addr;
+
+	coretalk_addr = PHYS_TO_TIODMA(page_addr);
+	if (!coretalk_addr) {
+		return 0;
+	}
+
+	return tioca_paddr_to_gart(coretalk_addr);
+}
+
+/**
+ * tioca_tlbflush - invalidate cached SGI CA GART TLB entries
+ * @tioca_kernel: CA context 
+ *
+ * Invalidate tlb entries for a given CA GART.  Main complexity is to account
+ * for revA bug.
+ */
+static inline void
+tioca_tlbflush(struct tioca_kernel *tioca_kernel)
+{
+	volatile uint64_t tmp;
+	volatile struct tioca *ca_base;
+	struct tioca_common *tioca_common;
+
+	tioca_common = tioca_kernel->ca_common;
+	ca_base = (struct tioca *)tioca_common->ca_common.bs_base;
+
+	/*
+	 * Explicit flushes not needed if GART is in cached mode
+	 */
+	if (tioca_kernel->ca_gart_iscoherent) {
+		if (TIOCA_WAR_ENABLED(PV910244, tioca_common)) {
+			/*
+			 * PV910244:  RevA CA needs explicit flushes.
+			 * Need to put GART into uncached mode before
+			 * flushing otherwise the explicit flush is ignored.
+			 *
+			 * Alternate WAR would be to leave GART cached and
+			 * touch every CL aligned GART entry.
+			 */
+
+			ca_base->ca_control2 &= ~(CA_GART_MEM_PARAM);
+			ca_base->ca_control2 |= CA_GART_FLUSH_TLB;
+			ca_base->ca_control2 |=
+			    (0x2ull << CA_GART_MEM_PARAM_SHFT);
+			tmp = ca_base->ca_control2;
+		}
+
+		return;
+	}
+
+	/*
+	 * Gart in uncached mode ... need an explicit flush.
+	 */
+
+	ca_base->ca_control2 |= CA_GART_FLUSH_TLB;
+	tmp = ca_base->ca_control2;
+}
+
+extern uint32_t	tioca_gart_found;
+extern int tioca_init_provider(void);
+extern void tioca_fastwrite_enable(struct tioca_kernel *tioca_kern);
+#endif /* _ASM_IA64_SN_TIO_CA_AGP_PROVIDER_H */
diff --git a/include/asm-ia64/sn/tiocx.h b/include/asm-ia64/sn/tiocx.h
new file mode 100644
index 00000000000..c5447a50450
--- /dev/null
+++ b/include/asm-ia64/sn/tiocx.h
@@ -0,0 +1,71 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (c) 2005 Silicon Graphics, Inc. All rights reserved.
+ */
+
+#ifndef _ASM_IA64_SN_TIO_TIOCX_H
+#define _ASM_IA64_SN_TIO_TIOCX_H
+
+#ifdef __KERNEL__
+
+struct cx_id_s {
+	unsigned int part_num;
+	unsigned int mfg_num;
+	int nasid;
+};
+
+struct cx_dev {
+	struct cx_id_s cx_id;
+	void *soft;			/* driver specific */
+	struct hubdev_info *hubdev;
+	struct device dev;
+	struct cx_drv *driver;
+};
+
+struct cx_device_id {
+	unsigned int part_num;
+	unsigned int mfg_num;
+};
+
+struct cx_drv {
+	char *name;
+	const struct cx_device_id *id_table;
+	struct device_driver driver;
+	int (*probe) (struct cx_dev * dev, const struct cx_device_id * id);
+	int (*remove) (struct cx_dev * dev);
+};
+
+/* create DMA address by stripping AS bits */
+#define TIOCX_DMA_ADDR(a) (uint64_t)((uint64_t)(a) & 0xffffcfffffffffUL)
+
+#define TIOCX_TO_TIOCX_DMA_ADDR(a) (uint64_t)(((uint64_t)(a) & 0xfffffffff) |  \
+                                  ((((uint64_t)(a)) & 0xffffc000000000UL) <<2))
+
+#define TIO_CE_ASIC_PARTNUM 0xce00
+#define TIOCX_CORELET 3
+
+/* These are taken from tio_mmr_as.h */
+#define TIO_ICE_FRZ_CFG               TIO_MMR_ADDR_MOD(0x00000000b0008100UL)
+#define TIO_ICE_PMI_TX_CFG            TIO_MMR_ADDR_MOD(0x00000000b000b100UL)
+#define TIO_ICE_PMI_TX_DYN_CREDIT_STAT_CB3 TIO_MMR_ADDR_MOD(0x00000000b000be18UL)
+#define TIO_ICE_PMI_TX_DYN_CREDIT_STAT_CB3_CREDIT_CNT_MASK 0x000000000000000fUL
+
+#define to_cx_dev(n) container_of(n, struct cx_dev, dev)
+#define to_cx_driver(drv) container_of(drv, struct cx_drv, driver)
+
+extern struct sn_irq_info *tiocx_irq_alloc(nasid_t, int, int, nasid_t, int);
+extern void tiocx_irq_free(struct sn_irq_info *);
+extern int cx_device_unregister(struct cx_dev *);
+extern int cx_device_register(nasid_t, int, int, struct hubdev_info *);
+extern int cx_driver_unregister(struct cx_drv *);
+extern int cx_driver_register(struct cx_drv *);
+extern uint64_t tiocx_dma_addr(uint64_t addr);
+extern uint64_t tiocx_swin_base(int nasid);
+extern void tiocx_mmr_store(int nasid, uint64_t offset, uint64_t value);
+extern uint64_t tiocx_mmr_load(int nasid, uint64_t offset);
+
+#endif				//  __KERNEL__
+#endif				// _ASM_IA64_SN_TIO_TIOCX__
diff --git a/include/asm-ia64/sn/types.h b/include/asm-ia64/sn/types.h
index 586ed47cae9..8e04ee211e5 100644
--- a/include/asm-ia64/sn/types.h
+++ b/include/asm-ia64/sn/types.h
@@ -16,7 +16,8 @@ typedef signed short	nasid_t;	/* node id in numa-as-id space */
 typedef signed char	partid_t;	/* partition ID type */
 typedef unsigned int    moduleid_t;     /* user-visible module number type */
 typedef unsigned int    cmoduleid_t;    /* kernel compact module id type */
-typedef signed char     slabid_t;
+typedef unsigned char	slotid_t;	/* slot (blade) within module */
+typedef unsigned char	slabid_t;	/* slab (asic) within slot */
 typedef u64 nic_t;
 typedef unsigned long iopaddr_t;
 typedef unsigned long paddr_t;